|
Hey there, Google Chrome shipped WebMCP in early preview this week. It's a W3C standard that lets any website expose structured tools directly to AI agents through a browser API. No MCP server required. Platform shifts build on existing networks. What's more likely — every company builds their own MCP server, or agents get good at using the web that already exists? MCP adoption is growing fast but it's still a tiny fraction of companies. WebMCP bets on the existing network. Always bet on the web. A few other big moves this week: OpenAI acquired OpenClaw — Peter Steinberger joins to build "the next generation of personal agents." OpenClaw moves to an open-source foundation. 180K GitHub stars, fastest-growing repo in GitHub history. OSS Agent orchestration is valuable enough to acquire now. Anthropic raised $30B at $380B — I listened to Dario Amodei's latest podcast where he says we're "near the end of the exponential" in raw model capabilities. The interesting thing about exponentials with limits: they hit those limits very quickly. We'll know soon whether there are real limits stopping digital god, or we'll have AGI. Gemini 3 Deep Think got a major upgrade — 84.6% on ARC-AGI-2, 3455 Elo on Codeforces, gold-medal at IMO 2025, 48.4% on Humanity's Last Exam. API access is still behind an early access form — worth signing up if you're doing anything with complex reasoning. Worth a look: Mixedbread — upload your docs (PDFs, audio, video, whatever) and get a search API back. Done-for-you knowledge base with adapters across formats. That's all for now. Keep shipping. — Drew p.s. rumor mill says there might be a GPT model launch soon |
Making the web faster at ▲Vercel while bootstrapping SaaS on the side. Currently building tryslater.com ⏰
Hey there, Two new posts this week plus a bunch of stuff I've been thinking about. Two Waves of AI — Most people treat AI as one wave. It's actually two. Wave 1 is LLMs as cheap, scalable intelligence (AI added to products). Wave 2 is 10x productivity in verifiable domains like coding (AI added to the creation process). How to Break Into GTM Engineering — We're hiring at Vercel and I wrote up the trends I’m seeing. Four paths in: solutions engineers, former founders, product-obsessed...
Hey there, This week I’m sharing the full breakdown of how we built Vercel’s first GTM agent. The results: 10 inbound SDRs → 1, $2M+ in savings, and a 32x ROI. I joined Alex Lieberman on the Human in the Loop podcast to walk through the whole thing. The blog post is a cleaned-up version of that conversation with everything you need to build something similar. Building Vercel’s First GTM Agent The short version: we turned lead qualification from a slow slog into something that happens in...
Hey there, AI is here to stay as part of your workflow. Why isn't it showing up in productivity numbers like public company earnings, GDP, or number of Jira tickets completed in a sprint? It's a good question — my thoughts on it here: https://drew.tech/posts/dev-efficiency tldr: Sandbagging AI often writes sub-senior eng quality code Large code diffs are hard to reason about So what to do? Figure out how to get past these. Don't sandbag. Push AI to meet your quality bar with prompts, rules,...